Or: why knowing every new AI tool is the wrong flex
I was at Lowe’s last weekend staring at the oscillating multi-tool wall. Fourteen variants. Corded, cordless, brushless, brushed, three battery platforms, a “pro” line I’d never heard of, and one that apparently also does grout.
I own one oscillating tool and I’ve used it maybe six times. Every time I’ve used it, it was the right tool, and I was glad I had it.
That shelf is what AI feels like right now.
A new tool every two hours
Between Tuesday morning coffee and Tuesday afternoon standup, three new coding agents will ship, someone will post a benchmark that invalidates last week’s benchmark, and a VC on LinkedIn will tell you the model you picked on Monday is already legacy.
The reflex is to keep up, try each one, build an opinion and have a take ready for the next exec meeting.
That reflex is the trap. It’s the same trap as the guy at Lowe’s who owns every tool on the shelf and still can’t finish his deck.
The contractor test
Here’s what I notice about contractors who actually build things for a living. They don’t own every tool. They own the tools that match the jobs they take. A framer’s truck looks nothing like a finish carpenter’s truck. A plumber’s van looks nothing like an electrician’s van. And none of them are anxious about the new Milwaukee release because they already know whether it solves a problem they’re actually being paid to solve.
The homeowner is the one with tool anxiety. The homeowner buys the shiny thing, uses it once, and it lives in the garage.
Right now, most engineering teams are buying AI tools like homeowners.
Enjoying this? I write about AI implementation and engineering leadership every week.
Stop tool-shopping. Start job-shopping.
The question isn’t “have I tried the new thing.” The question is “what job am I trying to finish, and is the tool I already have good enough to finish it.”
I wrote about this from a different angle in Minutes Added to Workforce. The real unit of AI value isn’t the tool, it’s the capacity it gives back to the humans doing the work. Capacity is measured against a job & tools without a job are cosplay.
So before the next team-wide evaluation of the next shiny agent, try this instead:
Write down the three jobs your team is losing sleep over this quarter. “Cut PR review time in half.” “Get incident MTTR under 20 minutes.” “Ship the migration without pulling two engineers off roadmap.”
Then ask which tool on the shelf actually addresses one of those three. If nothing on the current shelf beats what you already have, the answer is keep working. That is a valid, disciplined, senior answer.
The cognitive overhead tax nobody is pricing in
Every new tool your team adopts has a hidden bill: onboarding, context-switching, chat thread debating whether we should migrate, the one engineer who becomes the internal expert and is now a single point of failure. The quiet tax on focus every time someone has to decide which tool to reach for.
I call this the cognitive overhead tax, and it’s the thing most AI adoption decks forget to model. The ROI slide shows tokens saved. It never shows the Tuesday afternoon your staff engineer spent comparing three frameworks instead of shipping.
A shelf with 14 oscillating tools isn’t a capability. Teams with one well-chosen tool they’ve mastered consistently out-ship teams drowning in options.
What I actually do
I’ve watched my own team cycle through the hype, and my rule has gotten more boring over time, not less:
- Name the job first. Specific, measurable, this-quarter.
- Try the incumbent first. The tool you already know usually gets you 70% there and you’ll ship faster than the team doing the bake-off.
- Only evaluate new tools against a real job with a real deadline. Benchmarks on X don’t count.
- Kill tools aggressively. If something isn’t in the top three jobs, it comes off the shelf.
The actual flex
The flex isn’t knowing every tool… it is finishing the deck.
Next time a new Agent/Skill/Tool/LLM brain/model drops at 2pm and your feed lights up, ask yourself which of your three jobs it actually helps with. If the answer is none, close the tab and go ship the thing you were already shipping.
Don’t be ignorant to how the world is moving, but without it distracting you. Your job is to deliver value, not deliver shiny tool implementations.